Whamcloud - gitweb
LU-6391 llite: Add client mount opt to ignore suppress_pings
[fs/lustre-release.git] / lustre / llite / llite_lib.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2015, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/llite/llite_lib.c
37  *
38  * Lustre Light Super operations
39  */
40
41 #define DEBUG_SUBSYSTEM S_LLITE
42
43 #include <linux/module.h>
44 #include <linux/statfs.h>
45 #include <linux/types.h>
46 #include <linux/version.h>
47 #include <linux/mm.h>
48 #include <linux/user_namespace.h>
49 #ifdef HAVE_UIDGID_HEADER
50 # include <linux/uidgid.h>
51 #endif
52
53 #include <lustre_ioctl.h>
54 #include <lustre_ha.h>
55 #include <lustre_dlm.h>
56 #include <lprocfs_status.h>
57 #include <lustre_disk.h>
58 #include <lustre_param.h>
59 #include <lustre_log.h>
60 #include <cl_object.h>
61 #include <obd_cksum.h>
62 #include "llite_internal.h"
63
64 struct kmem_cache *ll_file_data_slab;
65
66 #ifndef log2
67 #define log2(n) ffz(~(n))
68 #endif
69
70 static struct ll_sb_info *ll_init_sbi(void)
71 {
72         struct ll_sb_info *sbi = NULL;
73         unsigned long pages;
74         unsigned long lru_page_max;
75         struct sysinfo si;
76         class_uuid_t uuid;
77         int i;
78         ENTRY;
79
80         OBD_ALLOC_PTR(sbi);
81         if (sbi == NULL)
82                 RETURN(NULL);
83
84         spin_lock_init(&sbi->ll_lock);
85         mutex_init(&sbi->ll_lco.lco_lock);
86         spin_lock_init(&sbi->ll_pp_extent_lock);
87         spin_lock_init(&sbi->ll_process_lock);
88         sbi->ll_rw_stats_on = 0;
89
90         si_meminfo(&si);
91         pages = si.totalram - si.totalhigh;
92         lru_page_max = pages / 2;
93
94         /* initialize ll_cache data */
95         sbi->ll_cache = cl_cache_init(lru_page_max);
96         if (sbi->ll_cache == NULL) {
97                 OBD_FREE(sbi, sizeof(*sbi));
98                 RETURN(NULL);
99         }
100
101         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
102                                            SBI_DEFAULT_READAHEAD_MAX);
103         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
104         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
105                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
106         INIT_LIST_HEAD(&sbi->ll_conn_chain);
107         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
108
109         ll_generate_random_uuid(uuid);
110         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
111         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
112
113         sbi->ll_flags |= LL_SBI_VERBOSE;
114 #ifdef ENABLE_CHECKSUM
115         sbi->ll_flags |= LL_SBI_CHECKSUM;
116 #endif
117
118 #ifdef HAVE_LRU_RESIZE_SUPPORT
119         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
120 #endif
121
122         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
123                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
124                                pp_r_hist.oh_lock);
125                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
126                                pp_w_hist.oh_lock);
127         }
128
129         /* metadata statahead is enabled by default */
130         sbi->ll_sa_max = LL_SA_RPC_DEF;
131         atomic_set(&sbi->ll_sa_total, 0);
132         atomic_set(&sbi->ll_sa_wrong, 0);
133         atomic_set(&sbi->ll_sa_running, 0);
134         atomic_set(&sbi->ll_agl_total, 0);
135         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
136
137         /* root squash */
138         sbi->ll_squash.rsi_uid = 0;
139         sbi->ll_squash.rsi_gid = 0;
140         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
141         init_rwsem(&sbi->ll_squash.rsi_sem);
142
143         RETURN(sbi);
144 }
145
146 static void ll_free_sbi(struct super_block *sb)
147 {
148         struct ll_sb_info *sbi = ll_s2sbi(sb);
149         ENTRY;
150
151         if (sbi != NULL) {
152                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
153                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
154                 if (sbi->ll_cache != NULL) {
155                         cl_cache_decref(sbi->ll_cache);
156                         sbi->ll_cache = NULL;
157                 }
158                 OBD_FREE(sbi, sizeof(*sbi));
159         }
160         EXIT;
161 }
162
163 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
164                                     struct vfsmount *mnt)
165 {
166         struct inode *root = NULL;
167         struct ll_sb_info *sbi = ll_s2sbi(sb);
168         struct obd_device *obd;
169         struct obd_statfs *osfs = NULL;
170         struct ptlrpc_request *request = NULL;
171         struct obd_connect_data *data = NULL;
172         struct obd_uuid *uuid;
173         struct md_op_data *op_data;
174         struct lustre_md lmd;
175         u64 valid;
176         int size, err, checksum;
177         ENTRY;
178
179         obd = class_name2obd(md);
180         if (!obd) {
181                 CERROR("MD %s: not setup or attached\n", md);
182                 RETURN(-EINVAL);
183         }
184
185         OBD_ALLOC_PTR(data);
186         if (data == NULL)
187                 RETURN(-ENOMEM);
188
189         OBD_ALLOC_PTR(osfs);
190         if (osfs == NULL) {
191                 OBD_FREE_PTR(data);
192                 RETURN(-ENOMEM);
193         }
194
195         /* indicate the features supported by this client */
196         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
197                                   OBD_CONNECT_ATTRFID  |
198                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
199                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
200                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
201                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
202                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
203                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
204                                   OBD_CONNECT_EINPROGRESS |
205                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
206                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
207                                   OBD_CONNECT_MAX_EASIZE |
208                                   OBD_CONNECT_FLOCK_DEAD |
209                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
210                                   OBD_CONNECT_OPEN_BY_FID |
211                                   OBD_CONNECT_DIR_STRIPE |
212                                   OBD_CONNECT_BULK_MBITS;
213
214 #ifdef HAVE_LRU_RESIZE_SUPPORT
215         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
216                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
217 #endif
218 #ifdef CONFIG_FS_POSIX_ACL
219         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
220 #endif
221
222         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
223                 /* flag mdc connection as lightweight, only used for test
224                  * purpose, use with care */
225                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
226
227         data->ocd_ibits_known = MDS_INODELOCK_FULL;
228         data->ocd_version = LUSTRE_VERSION_CODE;
229
230         if (sb->s_flags & MS_RDONLY)
231                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
232         if (sbi->ll_flags & LL_SBI_USER_XATTR)
233                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
234
235 #ifdef HAVE_MS_FLOCK_LOCK
236         /* force vfs to use lustre handler for flock() calls - bug 10743 */
237         sb->s_flags |= MS_FLOCK_LOCK;
238 #endif
239 #ifdef MS_HAS_NEW_AOPS
240         sb->s_flags |= MS_HAS_NEW_AOPS;
241 #endif
242
243         if (sbi->ll_flags & LL_SBI_FLOCK)
244                 sbi->ll_fop = &ll_file_operations_flock;
245         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
246                 sbi->ll_fop = &ll_file_operations;
247         else
248                 sbi->ll_fop = &ll_file_operations_noflock;
249
250         /* real client */
251         data->ocd_connect_flags |= OBD_CONNECT_REAL;
252         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
253                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
254
255         /* always ping even if server suppress_pings */
256         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
257                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
258
259         data->ocd_brw_size = MD_MAX_BRW_SIZE;
260
261         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
262         if (err == -EBUSY) {
263                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
264                                    "recovery, of which this client is not a "
265                                    "part. Please wait for recovery to complete,"
266                                    " abort, or time out.\n", md);
267                 GOTO(out, err);
268         } else if (err) {
269                 CERROR("cannot connect to %s: rc = %d\n", md, err);
270                 GOTO(out, err);
271         }
272
273         sbi->ll_md_exp->exp_connect_data = *data;
274
275         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
276                            LUSTRE_SEQ_METADATA);
277         if (err) {
278                 CERROR("%s: Can't init metadata layer FID infrastructure, "
279                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
280                 GOTO(out_md, err);
281         }
282
283         /* For mount, we only need fs info from MDT0, and also in DNE, it
284          * can make sure the client can be mounted as long as MDT0 is
285          * avaible */
286         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
287                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
288                         OBD_STATFS_FOR_MDT0);
289         if (err)
290                 GOTO(out_md_fid, err);
291
292         /* This needs to be after statfs to ensure connect has finished.
293          * Note that "data" does NOT contain the valid connect reply.
294          * If connecting to a 1.8 server there will be no LMV device, so
295          * we can access the MDC export directly and exp_connect_flags will
296          * be non-zero, but if accessing an upgraded 2.1 server it will
297          * have the correct flags filled in.
298          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
299         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
300         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
301             valid != CLIENT_CONNECT_MDT_REQD) {
302                 char *buf;
303
304                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
305                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
306                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
307                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
308                                    "feature(s) needed for correct operation "
309                                    "of this client (%s). Please upgrade "
310                                    "server or downgrade client.\n",
311                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
312                 OBD_FREE(buf, PAGE_CACHE_SIZE);
313                 GOTO(out_md_fid, err = -EPROTO);
314         }
315
316         size = sizeof(*data);
317         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
318                            KEY_CONN_DATA,  &size, data);
319         if (err) {
320                 CERROR("%s: Get connect data failed: rc = %d\n",
321                        sbi->ll_md_exp->exp_obd->obd_name, err);
322                 GOTO(out_md_fid, err);
323         }
324
325         LASSERT(osfs->os_bsize);
326         sb->s_blocksize = osfs->os_bsize;
327         sb->s_blocksize_bits = log2(osfs->os_bsize);
328         sb->s_magic = LL_SUPER_MAGIC;
329         sb->s_maxbytes = MAX_LFS_FILESIZE;
330         sbi->ll_namelen = osfs->os_namelen;
331
332         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
333             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
334                 LCONSOLE_INFO("Disabling user_xattr feature because "
335                               "it is not supported on the server\n");
336                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
337         }
338
339         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
340 #ifdef MS_POSIXACL
341                 sb->s_flags |= MS_POSIXACL;
342 #endif
343                 sbi->ll_flags |= LL_SBI_ACL;
344         } else {
345                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
346 #ifdef MS_POSIXACL
347                 sb->s_flags &= ~MS_POSIXACL;
348 #endif
349                 sbi->ll_flags &= ~LL_SBI_ACL;
350         }
351
352         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
353                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
354                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
355                         LCONSOLE_INFO("client is set as remote by default.\n");
356                 }
357         } else {
358                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
359                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
360                         LCONSOLE_INFO("client claims to be remote, but server "
361                                       "rejected, forced to be local.\n");
362                 }
363         }
364
365         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
366                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
367
368         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
369                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
370         else
371                 sbi->ll_md_brw_pages = 1;
372
373         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
374                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
375
376         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
377                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
378                         LCONSOLE_INFO("%s: disabling xattr cache due to "
379                                       "unknown maximum xattr size.\n", dt);
380                 } else {
381                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
382                         sbi->ll_xattr_cache_enabled = 1;
383                 }
384         }
385
386         obd = class_name2obd(dt);
387         if (!obd) {
388                 CERROR("DT %s: not setup or attached\n", dt);
389                 GOTO(out_md_fid, err = -ENODEV);
390         }
391
392         /* pass client page size via ocd_grant_blkbits, the server should report
393          * back its backend blocksize for grant calculation purpose */
394         data->ocd_grant_blkbits = PAGE_SHIFT;
395
396         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
397                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
398                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
399                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
400                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
401                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
402                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
403                                   OBD_CONNECT_MAXBYTES |
404                                   OBD_CONNECT_EINPROGRESS |
405                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
406                                   OBD_CONNECT_LAYOUTLOCK |
407                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
408                                   OBD_CONNECT_BULK_MBITS;
409
410         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
411                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
412
413         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
414                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
415                  * disabled by default, because it can still be enabled on the
416                  * fly via /proc. As a consequence, we still need to come to an
417                  * agreement on the supported algorithms at connect time */
418                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
419
420                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
421                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
422                 else
423                         data->ocd_cksum_types = cksum_types_supported_client();
424         }
425
426 #ifdef HAVE_LRU_RESIZE_SUPPORT
427         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
428 #endif
429         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
430                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
431
432         /* always ping even if server suppress_pings */
433         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
434                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
435
436         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
437                "ocd_grant: %d\n", data->ocd_connect_flags,
438                data->ocd_version, data->ocd_grant);
439
440         obd->obd_upcall.onu_owner = &sbi->ll_lco;
441         obd->obd_upcall.onu_upcall = cl_ocd_update;
442
443         data->ocd_brw_size = DT_MAX_BRW_SIZE;
444
445         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
446                           NULL);
447         if (err == -EBUSY) {
448                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
449                                    "recovery, of which this client is not a "
450                                    "part.  Please wait for recovery to "
451                                    "complete, abort, or time out.\n", dt);
452                 GOTO(out_md, err);
453         } else if (err) {
454                 CERROR("%s: Cannot connect to %s: rc = %d\n",
455                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
456                 GOTO(out_md, err);
457         }
458
459         sbi->ll_dt_exp->exp_connect_data = *data;
460
461         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
462                            LUSTRE_SEQ_METADATA);
463         if (err) {
464                 CERROR("%s: Can't init data layer FID infrastructure, "
465                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
466                 GOTO(out_dt, err);
467         }
468
469         mutex_lock(&sbi->ll_lco.lco_lock);
470         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
471         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
472         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
473         mutex_unlock(&sbi->ll_lco.lco_lock);
474
475         fid_zero(&sbi->ll_root_fid);
476         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid);
477         if (err) {
478                 CERROR("cannot mds_connect: rc = %d\n", err);
479                 GOTO(out_lock_cn_cb, err);
480         }
481         if (!fid_is_sane(&sbi->ll_root_fid)) {
482                 CERROR("%s: Invalid root fid "DFID" during mount\n",
483                        sbi->ll_md_exp->exp_obd->obd_name,
484                        PFID(&sbi->ll_root_fid));
485                 GOTO(out_lock_cn_cb, err = -EINVAL);
486         }
487         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
488
489         sb->s_op = &lustre_super_operations;
490 #if THREAD_SIZE >= 8192 /*b=17630*/
491         sb->s_export_op = &lustre_export_operations;
492 #endif
493
494         /* make root inode
495          * XXX: move this to after cbd setup? */
496         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
497         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
498                 valid |= OBD_MD_FLRMTPERM;
499         else if (sbi->ll_flags & LL_SBI_ACL)
500                 valid |= OBD_MD_FLACL;
501
502         OBD_ALLOC_PTR(op_data);
503         if (op_data == NULL)
504                 GOTO(out_lock_cn_cb, err = -ENOMEM);
505
506         op_data->op_fid1 = sbi->ll_root_fid;
507         op_data->op_mode = 0;
508         op_data->op_valid = valid;
509
510         err = md_getattr(sbi->ll_md_exp, op_data, &request);
511
512         OBD_FREE_PTR(op_data);
513         if (err) {
514                 CERROR("%s: md_getattr failed for root: rc = %d\n",
515                        sbi->ll_md_exp->exp_obd->obd_name, err);
516                 GOTO(out_lock_cn_cb, err);
517         }
518
519         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
520                                sbi->ll_md_exp, &lmd);
521         if (err) {
522                 CERROR("failed to understand root inode md: rc = %d\n", err);
523                 ptlrpc_req_finished(request);
524                 GOTO(out_lock_cn_cb, err);
525         }
526
527         LASSERT(fid_is_sane(&sbi->ll_root_fid));
528         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
529                                             sbi->ll_flags & LL_SBI_32BIT_API),
530                        &lmd);
531         md_free_lustre_md(sbi->ll_md_exp, &lmd);
532         ptlrpc_req_finished(request);
533
534         if (IS_ERR(root)) {
535 #ifdef CONFIG_FS_POSIX_ACL
536                 if (lmd.posix_acl) {
537                         posix_acl_release(lmd.posix_acl);
538                         lmd.posix_acl = NULL;
539                 }
540 #endif
541                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
542                 root = NULL;
543                 CERROR("lustre_lite: bad iget4 for root\n");
544                 GOTO(out_root, err);
545         }
546
547 #ifdef CONFIG_FS_POSIX_ACL
548         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
549                 rct_init(&sbi->ll_rct);
550                 et_init(&sbi->ll_et);
551         }
552 #endif
553
554         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
555         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
556                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
557                                  NULL);
558         cl_sb_init(sb);
559
560         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
561                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
562                                  sbi->ll_cache, NULL);
563
564         sb->s_root = d_make_root(root);
565         if (sb->s_root == NULL) {
566                 CERROR("%s: can't make root dentry\n",
567                         ll_get_fsname(sb, NULL, 0));
568                 GOTO(out_root, err = -ENOMEM);
569         }
570 #ifdef HAVE_DCACHE_LOCK
571         sb->s_root->d_op = &ll_d_ops;
572 #endif
573
574         sbi->ll_sdev_orig = sb->s_dev;
575
576         /* We set sb->s_dev equal on all lustre clients in order to support
577          * NFS export clustering.  NFSD requires that the FSID be the same
578          * on all clients. */
579         /* s_dev is also used in lt_compare() to compare two fs, but that is
580          * only a node-local comparison. */
581         uuid = obd_get_uuid(sbi->ll_md_exp);
582         if (uuid != NULL)
583                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
584
585         if (data != NULL)
586                 OBD_FREE_PTR(data);
587         if (osfs != NULL)
588                 OBD_FREE_PTR(osfs);
589         if (proc_lustre_fs_root != NULL) {
590                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
591                                                   dt, md);
592                 if (err < 0) {
593                         CERROR("%s: could not register mount in lprocfs: "
594                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
595                         err = 0;
596                 }
597         }
598
599         RETURN(err);
600 out_root:
601         if (root)
602                 iput(root);
603 out_lock_cn_cb:
604         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
605 out_dt:
606         obd_disconnect(sbi->ll_dt_exp);
607         sbi->ll_dt_exp = NULL;
608 out_md_fid:
609         obd_fid_fini(sbi->ll_md_exp->exp_obd);
610 out_md:
611         obd_disconnect(sbi->ll_md_exp);
612         sbi->ll_md_exp = NULL;
613 out:
614         if (data != NULL)
615                 OBD_FREE_PTR(data);
616         if (osfs != NULL)
617                 OBD_FREE_PTR(osfs);
618         return err;
619 }
620
621 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
622 {
623         int size, rc;
624
625         size = sizeof(*lmmsize);
626         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
627                           KEY_MAX_EASIZE, &size, lmmsize);
628         if (rc != 0) {
629                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
630                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
631                 RETURN(rc);
632         }
633
634         size = sizeof(int);
635         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
636                           KEY_MAX_EASIZE, &size, lmmsize);
637         if (rc)
638                 CERROR("Get max mdsize error rc %d\n", rc);
639
640         RETURN(rc);
641 }
642
643 /**
644  * Get the value of the default_easize parameter.
645  *
646  * \see client_obd::cl_default_mds_easize
647  *
648  * \param[in] sbi       superblock info for this filesystem
649  * \param[out] lmmsize  pointer to storage location for value
650  *
651  * \retval 0            on success
652  * \retval negative     negated errno on failure
653  */
654 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
655 {
656         int size, rc;
657
658         size = sizeof(int);
659         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
660                          KEY_DEFAULT_EASIZE, &size, lmmsize);
661         if (rc)
662                 CERROR("Get default mdsize error rc %d\n", rc);
663
664         RETURN(rc);
665 }
666
667 /**
668  * Set the default_easize parameter to the given value.
669  *
670  * \see client_obd::cl_default_mds_easize
671  *
672  * \param[in] sbi       superblock info for this filesystem
673  * \param[in] lmmsize   the size to set
674  *
675  * \retval 0            on success
676  * \retval negative     negated errno on failure
677  */
678 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
679 {
680         int rc;
681
682         if (lmmsize < sizeof(struct lov_mds_md) ||
683             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
684                 return -EINVAL;
685
686         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
687                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
688                                 sizeof(int), &lmmsize, NULL);
689
690         RETURN(rc);
691 }
692
693 static void ll_dump_inode(struct inode *inode)
694 {
695         struct ll_d_hlist_node *tmp;
696         int dentry_count = 0;
697
698         LASSERT(inode != NULL);
699
700         ll_d_hlist_for_each(tmp, &inode->i_dentry)
701                 dentry_count++;
702
703         CERROR("%s: inode %p dump: dev=%s fid="DFID
704                " mode=%o count=%u, %d dentries\n",
705                ll_get_fsname(inode->i_sb, NULL, 0), inode,
706                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
707                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
708 }
709
710 void lustre_dump_dentry(struct dentry *dentry, int recur)
711 {
712         struct list_head *tmp;
713         int subdirs = 0;
714
715         LASSERT(dentry != NULL);
716
717         list_for_each(tmp, &dentry->d_subdirs)
718                 subdirs++;
719
720         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
721                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
722                dentry->d_name.len, dentry->d_name.name,
723                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
724                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
725                dentry->d_flags, dentry->d_fsdata, subdirs);
726         if (dentry->d_inode != NULL)
727                 ll_dump_inode(dentry->d_inode);
728
729         if (recur == 0)
730                 return;
731
732         list_for_each(tmp, &dentry->d_subdirs) {
733                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
734                 lustre_dump_dentry(d, recur - 1);
735         }
736 }
737
738 static void client_common_put_super(struct super_block *sb)
739 {
740         struct ll_sb_info *sbi = ll_s2sbi(sb);
741         ENTRY;
742
743 #ifdef CONFIG_FS_POSIX_ACL
744         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
745                 et_fini(&sbi->ll_et);
746                 rct_fini(&sbi->ll_rct);
747         }
748 #endif
749
750         cl_sb_fini(sb);
751
752         list_del(&sbi->ll_conn_chain);
753
754         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
755         obd_disconnect(sbi->ll_dt_exp);
756         sbi->ll_dt_exp = NULL;
757
758         lprocfs_unregister_mountpoint(sbi);
759
760         obd_fid_fini(sbi->ll_md_exp->exp_obd);
761         obd_disconnect(sbi->ll_md_exp);
762         sbi->ll_md_exp = NULL;
763
764         EXIT;
765 }
766
767 void ll_kill_super(struct super_block *sb)
768 {
769         struct ll_sb_info *sbi;
770         ENTRY;
771
772         /* not init sb ?*/
773         if (!(sb->s_flags & MS_ACTIVE))
774                 return;
775
776         sbi = ll_s2sbi(sb);
777         /* we need restore s_dev from changed for clustred NFS before put_super
778          * because new kernels have cached s_dev and change sb->s_dev in
779          * put_super not affected real removing devices */
780         if (sbi) {
781                 sb->s_dev = sbi->ll_sdev_orig;
782                 sbi->ll_umounting = 1;
783
784                 /* wait running statahead threads to quit */
785                 while (atomic_read(&sbi->ll_sa_running) > 0) {
786                         set_current_state(TASK_UNINTERRUPTIBLE);
787                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
788                 }
789         }
790
791         EXIT;
792 }
793
794 static inline int ll_set_opt(const char *opt, char *data, int fl)
795 {
796         if (strncmp(opt, data, strlen(opt)) != 0)
797                 return(0);
798         else
799                 return(fl);
800 }
801
802 /* non-client-specific mount options are parsed in lmd_parse */
803 static int ll_options(char *options, int *flags)
804 {
805         int tmp;
806         char *s1 = options, *s2;
807         ENTRY;
808
809         if (!options)
810                 RETURN(0);
811
812         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
813
814         while (*s1) {
815                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
816                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
817                 if (tmp) {
818                         *flags |= tmp;
819                         goto next;
820                 }
821                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
822                 if (tmp) {
823                         *flags |= tmp;
824                         goto next;
825                 }
826                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
827                 if (tmp) {
828                         *flags |= tmp;
829                         goto next;
830                 }
831                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
832                 if (tmp) {
833                         *flags &= ~tmp;
834                         goto next;
835                 }
836                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
837                 if (tmp) {
838                         *flags |= tmp;
839                         goto next;
840                 }
841                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
842                 if (tmp) {
843                         *flags &= ~tmp;
844                         goto next;
845                 }
846                 tmp = ll_set_opt("context", s1, 1);
847                 if (tmp)
848                         goto next;
849                 tmp = ll_set_opt("fscontext", s1, 1);
850                 if (tmp)
851                         goto next;
852                 tmp = ll_set_opt("defcontext", s1, 1);
853                 if (tmp)
854                         goto next;
855                 tmp = ll_set_opt("rootcontext", s1, 1);
856                 if (tmp)
857                         goto next;
858                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
859                 if (tmp) {
860                         *flags |= tmp;
861                         goto next;
862                 }
863                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
864                 if (tmp) {
865                         *flags |= tmp;
866                         goto next;
867                 }
868                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
869                 if (tmp) {
870                         *flags &= ~tmp;
871                         goto next;
872                 }
873
874                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
875                 if (tmp) {
876                         *flags |= tmp;
877                         goto next;
878                 }
879                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
880                 if (tmp) {
881                         *flags &= ~tmp;
882                         goto next;
883                 }
884                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
885                 if (tmp) {
886                         *flags |= tmp;
887                         goto next;
888                 }
889                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
890                 if (tmp) {
891                         *flags &= ~tmp;
892                         goto next;
893                 }
894                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
895                 if (tmp) {
896                         *flags |= tmp;
897                         goto next;
898                 }
899                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
900                 if (tmp) {
901                         *flags &= ~tmp;
902                         goto next;
903                 }
904                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
905                 if (tmp) {
906                         *flags |= tmp;
907                         goto next;
908                 }
909                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
910                 if (tmp) {
911                         *flags |= tmp;
912                         goto next;
913                 }
914                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
915                 if (tmp) {
916                         *flags &= ~tmp;
917                         goto next;
918                 }
919                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
920                 if (tmp) {
921                         *flags |= tmp;
922                         goto next;
923                 }
924                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
925                                    s1);
926                 RETURN(-EINVAL);
927
928 next:
929                 /* Find next opt */
930                 s2 = strchr(s1, ',');
931                 if (s2 == NULL)
932                         break;
933                 s1 = s2 + 1;
934         }
935         RETURN(0);
936 }
937
938 void ll_lli_init(struct ll_inode_info *lli)
939 {
940         lli->lli_inode_magic = LLI_INODE_MAGIC;
941         lli->lli_flags = 0;
942         spin_lock_init(&lli->lli_lock);
943         lli->lli_posix_acl = NULL;
944         lli->lli_remote_perms = NULL;
945         mutex_init(&lli->lli_rmtperm_mutex);
946         /* Do not set lli_fid, it has been initialized already. */
947         fid_zero(&lli->lli_pfid);
948         atomic_set(&lli->lli_open_count, 0);
949         lli->lli_rmtperm_time = 0;
950         lli->lli_mds_read_och = NULL;
951         lli->lli_mds_write_och = NULL;
952         lli->lli_mds_exec_och = NULL;
953         lli->lli_open_fd_read_count = 0;
954         lli->lli_open_fd_write_count = 0;
955         lli->lli_open_fd_exec_count = 0;
956         mutex_init(&lli->lli_och_mutex);
957         spin_lock_init(&lli->lli_agl_lock);
958         spin_lock_init(&lli->lli_layout_lock);
959         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
960         lli->lli_clob = NULL;
961
962         init_rwsem(&lli->lli_xattrs_list_rwsem);
963         mutex_init(&lli->lli_xattrs_enq_lock);
964
965         LASSERT(lli->lli_vfs_inode.i_mode != 0);
966         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
967                 mutex_init(&lli->lli_readdir_mutex);
968                 lli->lli_opendir_key = NULL;
969                 lli->lli_sai = NULL;
970                 spin_lock_init(&lli->lli_sa_lock);
971                 lli->lli_opendir_pid = 0;
972                 lli->lli_sa_enabled = 0;
973                 lli->lli_def_stripe_offset = -1;
974         } else {
975                 mutex_init(&lli->lli_size_mutex);
976                 lli->lli_symlink_name = NULL;
977                 init_rwsem(&lli->lli_trunc_sem);
978                 range_lock_tree_init(&lli->lli_write_tree);
979                 init_rwsem(&lli->lli_glimpse_sem);
980                 lli->lli_glimpse_time = 0;
981                 INIT_LIST_HEAD(&lli->lli_agl_list);
982                 lli->lli_agl_index = 0;
983                 lli->lli_async_rc = 0;
984         }
985         mutex_init(&lli->lli_layout_mutex);
986 }
987
988 static inline int ll_bdi_register(struct backing_dev_info *bdi)
989 {
990         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
991
992         bdi->name = "lustre";
993         return bdi_register(bdi, NULL, "lustre-%d",
994                             atomic_inc_return(&ll_bdi_num));
995 }
996
997 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
998 {
999         struct lustre_profile *lprof = NULL;
1000         struct lustre_sb_info *lsi = s2lsi(sb);
1001         struct ll_sb_info *sbi;
1002         char  *dt = NULL, *md = NULL;
1003         char  *profilenm = get_profile_name(sb);
1004         struct config_llog_instance *cfg;
1005         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1006         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1007         int    err;
1008         ENTRY;
1009
1010         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1011
1012         OBD_ALLOC_PTR(cfg);
1013         if (cfg == NULL)
1014                 RETURN(-ENOMEM);
1015
1016         try_module_get(THIS_MODULE);
1017
1018         /* client additional sb info */
1019         lsi->lsi_llsbi = sbi = ll_init_sbi();
1020         if (!sbi) {
1021                 module_put(THIS_MODULE);
1022                 OBD_FREE_PTR(cfg);
1023                 RETURN(-ENOMEM);
1024         }
1025
1026         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1027         if (err)
1028                 GOTO(out_free, err);
1029
1030         err = bdi_init(&lsi->lsi_bdi);
1031         if (err)
1032                 GOTO(out_free, err);
1033         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1034 #ifdef HAVE_BDI_CAP_MAP_COPY
1035         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1036 #else
1037         lsi->lsi_bdi.capabilities = 0;
1038 #endif
1039         err = ll_bdi_register(&lsi->lsi_bdi);
1040         if (err)
1041                 GOTO(out_free, err);
1042
1043         sb->s_bdi = &lsi->lsi_bdi;
1044 #ifndef HAVE_DCACHE_LOCK
1045         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1046         sb->s_d_op = &ll_d_ops;
1047 #endif
1048
1049         /* Generate a string unique to this super, in case some joker tries
1050            to mount the same fs at two mount points.
1051            Use the address of the super itself.*/
1052         cfg->cfg_instance = sb;
1053         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1054         cfg->cfg_callback = class_config_llog_handler;
1055         /* set up client obds */
1056         err = lustre_process_log(sb, profilenm, cfg);
1057         if (err < 0)
1058                 GOTO(out_free, err);
1059
1060         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1061         lprof = class_get_profile(profilenm);
1062         if (lprof == NULL) {
1063                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1064                                    " read from the MGS.  Does that filesystem "
1065                                    "exist?\n", profilenm);
1066                 GOTO(out_free, err = -EINVAL);
1067         }
1068         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1069                lprof->lp_md, lprof->lp_dt);
1070
1071         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1072         if (!dt)
1073                 GOTO(out_free, err = -ENOMEM);
1074         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1075
1076         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1077         if (!md)
1078                 GOTO(out_free, err = -ENOMEM);
1079         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1080
1081         /* connections, registrations, sb setup */
1082         err = client_common_fill_super(sb, md, dt, mnt);
1083         if (err < 0)
1084                 GOTO(out_free, err);
1085
1086         sbi->ll_client_common_fill_super_succeeded = 1;
1087
1088 out_free:
1089         if (md)
1090                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1091         if (dt)
1092                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1093         if (lprof != NULL)
1094                 class_put_profile(lprof);
1095         if (err)
1096                 ll_put_super(sb);
1097         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1098                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1099
1100         OBD_FREE_PTR(cfg);
1101         RETURN(err);
1102 } /* ll_fill_super */
1103
1104 void ll_put_super(struct super_block *sb)
1105 {
1106         struct config_llog_instance cfg, params_cfg;
1107         struct obd_device *obd;
1108         struct lustre_sb_info *lsi = s2lsi(sb);
1109         struct ll_sb_info *sbi = ll_s2sbi(sb);
1110         char *profilenm = get_profile_name(sb);
1111         long ccc_count;
1112         int next, force = 1, rc = 0;
1113         ENTRY;
1114
1115         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1116
1117         cfg.cfg_instance = sb;
1118         lustre_end_log(sb, profilenm, &cfg);
1119
1120         params_cfg.cfg_instance = sb;
1121         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1122
1123         if (sbi->ll_md_exp) {
1124                 obd = class_exp2obd(sbi->ll_md_exp);
1125                 if (obd)
1126                         force = obd->obd_force;
1127         }
1128
1129         /* Wait for unstable pages to be committed to stable storage */
1130         if (force == 0) {
1131                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1132                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1133                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1134                         &lwi);
1135         }
1136
1137         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1138         if (force == 0 && rc != -EINTR)
1139                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1140
1141
1142         /* We need to set force before the lov_disconnect in
1143            lustre_common_put_super, since l_d cleans up osc's as well. */
1144         if (force) {
1145                 next = 0;
1146                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1147                                                      &next)) != NULL) {
1148                         obd->obd_force = force;
1149                 }
1150         }
1151
1152         if (sbi->ll_client_common_fill_super_succeeded) {
1153                 /* Only if client_common_fill_super succeeded */
1154                 client_common_put_super(sb);
1155         }
1156
1157         next = 0;
1158         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1159                 class_manual_cleanup(obd);
1160         }
1161
1162         if (sbi->ll_flags & LL_SBI_VERBOSE)
1163                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1164
1165         if (profilenm)
1166                 class_del_profile(profilenm);
1167
1168         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1169                 bdi_destroy(&lsi->lsi_bdi);
1170                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1171         }
1172
1173         ll_free_sbi(sb);
1174         lsi->lsi_llsbi = NULL;
1175
1176         lustre_common_put_super(sb);
1177
1178         cl_env_cache_purge(~0);
1179
1180         module_put(THIS_MODULE);
1181
1182         EXIT;
1183 } /* client_put_super */
1184
1185 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1186 {
1187         struct inode *inode = NULL;
1188
1189         /* NOTE: we depend on atomic igrab() -bzzz */
1190         lock_res_and_lock(lock);
1191         if (lock->l_resource->lr_lvb_inode) {
1192                 struct ll_inode_info * lli;
1193                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1194                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1195                         inode = igrab(lock->l_resource->lr_lvb_inode);
1196                 } else {
1197                         inode = lock->l_resource->lr_lvb_inode;
1198                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1199                                          D_WARNING, lock, "lr_lvb_inode %p is "
1200                                          "bogus: magic %08x",
1201                                          lock->l_resource->lr_lvb_inode,
1202                                          lli->lli_inode_magic);
1203                         inode = NULL;
1204                 }
1205         }
1206         unlock_res_and_lock(lock);
1207         return inode;
1208 }
1209
1210 void ll_dir_clear_lsm_md(struct inode *inode)
1211 {
1212         struct ll_inode_info *lli = ll_i2info(inode);
1213
1214         LASSERT(S_ISDIR(inode->i_mode));
1215
1216         if (lli->lli_lsm_md != NULL) {
1217                 lmv_free_memmd(lli->lli_lsm_md);
1218                 lli->lli_lsm_md = NULL;
1219         }
1220 }
1221
1222 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1223                                       const struct lu_fid *fid,
1224                                       struct lustre_md *md)
1225 {
1226         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1227         struct mdt_body         *body = md->body;
1228         struct inode            *inode;
1229         ino_t                   ino;
1230         ENTRY;
1231
1232         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1233         inode = iget_locked(sb, ino);
1234         if (inode == NULL) {
1235                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1236                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1237                 RETURN(ERR_PTR(-ENOENT));
1238         }
1239
1240         if (inode->i_state & I_NEW) {
1241                 struct ll_inode_info *lli = ll_i2info(inode);
1242                 struct lmv_stripe_md *lsm = md->lmv;
1243
1244                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1245                                 (body->mbo_mode & S_IFMT);
1246                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1247                          PFID(fid));
1248
1249                 LTIME_S(inode->i_mtime) = 0;
1250                 LTIME_S(inode->i_atime) = 0;
1251                 LTIME_S(inode->i_ctime) = 0;
1252                 inode->i_rdev = 0;
1253
1254 #ifdef HAVE_BACKING_DEV_INFO
1255                 /* initializing backing dev info. */
1256                 inode->i_mapping->backing_dev_info =
1257                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1258 #endif
1259                 inode->i_op = &ll_dir_inode_operations;
1260                 inode->i_fop = &ll_dir_operations;
1261                 lli->lli_fid = *fid;
1262                 ll_lli_init(lli);
1263
1264                 LASSERT(lsm != NULL);
1265                 /* master object FID */
1266                 lli->lli_pfid = body->mbo_fid1;
1267                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1268                        lli, PFID(fid), PFID(&lli->lli_pfid));
1269                 unlock_new_inode(inode);
1270         }
1271
1272         RETURN(inode);
1273 }
1274
1275 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1276 {
1277         struct lu_fid *fid;
1278         struct lmv_stripe_md *lsm = md->lmv;
1279         int i;
1280
1281         LASSERT(lsm != NULL);
1282         /* XXX sigh, this lsm_root initialization should be in
1283          * LMV layer, but it needs ll_iget right now, so we
1284          * put this here right now. */
1285         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1286                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1287                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1288                 /* Unfortunately ll_iget will call ll_update_inode,
1289                  * where the initialization of slave inode is slightly
1290                  * different, so it reset lsm_md to NULL to avoid
1291                  * initializing lsm for slave inode. */
1292                 /* For migrating inode, master stripe and master object will
1293                  * be same, so we only need assign this inode */
1294                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1295                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1296                 else
1297                         lsm->lsm_md_oinfo[i].lmo_root =
1298                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1299
1300                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1301                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1302
1303                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1304                         return rc;
1305                 }
1306         }
1307
1308         return 0;
1309 }
1310
1311 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1312                                 const struct lmv_stripe_md *lsm_md2)
1313 {
1314         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1315                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1316                lsm_md1->lsm_md_master_mdt_index ==
1317                                         lsm_md2->lsm_md_master_mdt_index &&
1318                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1319                lsm_md1->lsm_md_layout_version ==
1320                                         lsm_md2->lsm_md_layout_version &&
1321                strcmp(lsm_md1->lsm_md_pool_name,
1322                       lsm_md2->lsm_md_pool_name) == 0;
1323 }
1324
1325 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1326 {
1327         struct ll_inode_info *lli = ll_i2info(inode);
1328         struct lmv_stripe_md *lsm = md->lmv;
1329         int     rc;
1330         ENTRY;
1331
1332         LASSERT(S_ISDIR(inode->i_mode));
1333         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1334                PFID(ll_inode2fid(inode)));
1335
1336         /* no striped information from request. */
1337         if (lsm == NULL) {
1338                 if (lli->lli_lsm_md == NULL) {
1339                         RETURN(0);
1340                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1341                                                 LMV_HASH_FLAG_MIGRATION) {
1342                         /* migration is done, the temporay MIGRATE layout has
1343                          * been removed */
1344                         CDEBUG(D_INODE, DFID" finish migration.\n",
1345                                PFID(ll_inode2fid(inode)));
1346                         lmv_free_memmd(lli->lli_lsm_md);
1347                         lli->lli_lsm_md = NULL;
1348                         RETURN(0);
1349                 } else {
1350                         /* The lustre_md from req does not include stripeEA,
1351                          * see ll_md_setattr */
1352                         RETURN(0);
1353                 }
1354         }
1355
1356         /* set the directory layout */
1357         if (lli->lli_lsm_md == NULL) {
1358                 struct cl_attr  *attr;
1359
1360                 rc = ll_init_lsm_md(inode, md);
1361                 if (rc != 0)
1362                         RETURN(rc);
1363
1364                 /* set md->lmv to NULL, so the following free lustre_md
1365                  * will not free this lsm */
1366                 md->lmv = NULL;
1367                 lli->lli_lsm_md = lsm;
1368
1369                 OBD_ALLOC_PTR(attr);
1370                 if (attr == NULL)
1371                         RETURN(-ENOMEM);
1372
1373                 /* validate the lsm */
1374                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1375                                    ll_md_blocking_ast);
1376                 if (rc != 0) {
1377                         OBD_FREE_PTR(attr);
1378                         RETURN(rc);
1379                 }
1380
1381                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1382                         md->body->mbo_nlink = attr->cat_nlink;
1383                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1384                         md->body->mbo_size = attr->cat_size;
1385                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1386                         md->body->mbo_atime = attr->cat_atime;
1387                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1388                         md->body->mbo_ctime = attr->cat_ctime;
1389                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1390                         md->body->mbo_mtime = attr->cat_mtime;
1391
1392                 OBD_FREE_PTR(attr);
1393
1394                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1395                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1396                 RETURN(0);
1397         }
1398
1399         /* Compare the old and new stripe information */
1400         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1401                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1402                 int                     idx;
1403
1404                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1405                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1406                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1407                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1408                        inode, lsm, old_lsm,
1409                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1410                        lsm->lsm_md_stripe_count,
1411                        old_lsm->lsm_md_stripe_count,
1412                        lsm->lsm_md_master_mdt_index,
1413                        old_lsm->lsm_md_master_mdt_index,
1414                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1415                        lsm->lsm_md_layout_version,
1416                        old_lsm->lsm_md_layout_version,
1417                        lsm->lsm_md_pool_name,
1418                        old_lsm->lsm_md_pool_name);
1419
1420                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1421                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1422                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1423                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1424                 }
1425
1426                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1427                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1428                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1429                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1430                 }
1431
1432                 RETURN(-EIO);
1433         }
1434
1435         RETURN(0);
1436 }
1437
1438 void ll_clear_inode(struct inode *inode)
1439 {
1440         struct ll_inode_info *lli = ll_i2info(inode);
1441         struct ll_sb_info *sbi = ll_i2sbi(inode);
1442         ENTRY;
1443
1444         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1445                PFID(ll_inode2fid(inode)), inode);
1446
1447         if (S_ISDIR(inode->i_mode)) {
1448                 /* these should have been cleared in ll_file_release */
1449                 LASSERT(lli->lli_opendir_key == NULL);
1450                 LASSERT(lli->lli_sai == NULL);
1451                 LASSERT(lli->lli_opendir_pid == 0);
1452         }
1453
1454         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1455
1456         LASSERT(!lli->lli_open_fd_write_count);
1457         LASSERT(!lli->lli_open_fd_read_count);
1458         LASSERT(!lli->lli_open_fd_exec_count);
1459
1460         if (lli->lli_mds_write_och)
1461                 ll_md_real_close(inode, FMODE_WRITE);
1462         if (lli->lli_mds_exec_och)
1463                 ll_md_real_close(inode, FMODE_EXEC);
1464         if (lli->lli_mds_read_och)
1465                 ll_md_real_close(inode, FMODE_READ);
1466
1467         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1468                 OBD_FREE(lli->lli_symlink_name,
1469                          strlen(lli->lli_symlink_name) + 1);
1470                 lli->lli_symlink_name = NULL;
1471         }
1472
1473         ll_xattr_cache_destroy(inode);
1474
1475         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1476                 LASSERT(lli->lli_posix_acl == NULL);
1477                 if (lli->lli_remote_perms) {
1478                         free_rmtperm_hash(lli->lli_remote_perms);
1479                         lli->lli_remote_perms = NULL;
1480                 }
1481         }
1482 #ifdef CONFIG_FS_POSIX_ACL
1483         else if (lli->lli_posix_acl) {
1484                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1485                 LASSERT(lli->lli_remote_perms == NULL);
1486                 posix_acl_release(lli->lli_posix_acl);
1487                 lli->lli_posix_acl = NULL;
1488         }
1489 #endif
1490         lli->lli_inode_magic = LLI_INODE_DEAD;
1491
1492         if (S_ISDIR(inode->i_mode))
1493                 ll_dir_clear_lsm_md(inode);
1494         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1495                 LASSERT(list_empty(&lli->lli_agl_list));
1496
1497         /*
1498          * XXX This has to be done before lsm is freed below, because
1499          * cl_object still uses inode lsm.
1500          */
1501         cl_inode_fini(inode);
1502
1503         EXIT;
1504 }
1505
1506 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1507 {
1508         struct lustre_md md;
1509         struct inode *inode = dentry->d_inode;
1510         struct ll_sb_info *sbi = ll_i2sbi(inode);
1511         struct ptlrpc_request *request = NULL;
1512         int rc, ia_valid;
1513         ENTRY;
1514
1515         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1516                                      LUSTRE_OPC_ANY, NULL);
1517         if (IS_ERR(op_data))
1518                 RETURN(PTR_ERR(op_data));
1519
1520         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1521         if (rc) {
1522                 ptlrpc_req_finished(request);
1523                 if (rc == -ENOENT) {
1524                         clear_nlink(inode);
1525                         /* Unlinked special device node? Or just a race?
1526                          * Pretend we done everything. */
1527                         if (!S_ISREG(inode->i_mode) &&
1528                             !S_ISDIR(inode->i_mode)) {
1529                                 ia_valid = op_data->op_attr.ia_valid;
1530                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1531                                 rc = simple_setattr(dentry, &op_data->op_attr);
1532                                 op_data->op_attr.ia_valid = ia_valid;
1533                         }
1534                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1535                         CERROR("md_setattr fails: rc = %d\n", rc);
1536                 }
1537                 RETURN(rc);
1538         }
1539
1540         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1541                               sbi->ll_md_exp, &md);
1542         if (rc) {
1543                 ptlrpc_req_finished(request);
1544                 RETURN(rc);
1545         }
1546
1547         ia_valid = op_data->op_attr.ia_valid;
1548         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1549          * cache is not cleared yet. */
1550         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1551         if (S_ISREG(inode->i_mode))
1552                 mutex_lock(&inode->i_mutex);
1553         rc = simple_setattr(dentry, &op_data->op_attr);
1554         if (S_ISREG(inode->i_mode))
1555                 mutex_unlock(&inode->i_mutex);
1556         op_data->op_attr.ia_valid = ia_valid;
1557
1558         rc = ll_update_inode(inode, &md);
1559         ptlrpc_req_finished(request);
1560
1561         RETURN(rc);
1562 }
1563
1564 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1565  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1566  * keep these values until such a time that objects are allocated for it.
1567  * We do the MDS operations first, as it is checking permissions for us.
1568  * We don't to the MDS RPC if there is nothing that we want to store there,
1569  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1570  * going to do an RPC anyways.
1571  *
1572  * If we are doing a truncate, we will send the mtime and ctime updates
1573  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1574  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1575  * at the same time.
1576  *
1577  * In case of HSMimport, we only set attr on MDS.
1578  */
1579 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1580 {
1581         struct inode *inode = dentry->d_inode;
1582         struct ll_inode_info *lli = ll_i2info(inode);
1583         struct md_op_data *op_data = NULL;
1584         int rc = 0;
1585         ENTRY;
1586
1587         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1588                "valid %x, hsm_import %d\n",
1589                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1590                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1591                hsm_import);
1592
1593         if (attr->ia_valid & ATTR_SIZE) {
1594                 /* Check new size against VFS/VM file size limit and rlimit */
1595                 rc = inode_newsize_ok(inode, attr->ia_size);
1596                 if (rc)
1597                         RETURN(rc);
1598
1599                 /* The maximum Lustre file size is variable, based on the
1600                  * OST maximum object size and number of stripes.  This
1601                  * needs another check in addition to the VFS check above. */
1602                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1603                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1604                                PFID(&lli->lli_fid), attr->ia_size,
1605                                ll_file_maxbytes(inode));
1606                         RETURN(-EFBIG);
1607                 }
1608
1609                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1610         }
1611
1612         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1613         if (attr->ia_valid & TIMES_SET_FLAGS) {
1614                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1615                     !cfs_capable(CFS_CAP_FOWNER))
1616                         RETURN(-EPERM);
1617         }
1618
1619         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1620         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1621             (attr->ia_valid & ATTR_CTIME)) {
1622                 attr->ia_ctime = CFS_CURRENT_TIME;
1623                 attr->ia_valid |= ATTR_CTIME_SET;
1624         }
1625         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1626             (attr->ia_valid & ATTR_ATIME)) {
1627                 attr->ia_atime = CFS_CURRENT_TIME;
1628                 attr->ia_valid |= ATTR_ATIME_SET;
1629         }
1630         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1631             (attr->ia_valid & ATTR_MTIME)) {
1632                 attr->ia_mtime = CFS_CURRENT_TIME;
1633                 attr->ia_valid |= ATTR_MTIME_SET;
1634         }
1635
1636         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1637                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1638                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1639                        cfs_time_current_sec());
1640
1641         if (S_ISREG(inode->i_mode)) {
1642                 if (attr->ia_valid & ATTR_SIZE)
1643                         inode_dio_write_done(inode);
1644                 mutex_unlock(&inode->i_mutex);
1645         }
1646
1647         /* We always do an MDS RPC, even if we're only changing the size;
1648          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1649
1650         OBD_ALLOC_PTR(op_data);
1651         if (op_data == NULL)
1652                 GOTO(out, rc = -ENOMEM);
1653
1654         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1655                 /* If we are changing file size, file content is
1656                  * modified, flag it. */
1657                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1658                 op_data->op_bias |= MDS_DATA_MODIFIED;
1659                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1660         }
1661
1662         op_data->op_attr = *attr;
1663
1664         rc = ll_md_setattr(dentry, op_data);
1665         if (rc)
1666                 GOTO(out, rc);
1667
1668         if (!S_ISREG(inode->i_mode) || hsm_import)
1669                 GOTO(out, rc = 0);
1670
1671         if (attr->ia_valid & (ATTR_SIZE |
1672                               ATTR_ATIME | ATTR_ATIME_SET |
1673                               ATTR_MTIME | ATTR_MTIME_SET |
1674                               ATTR_CTIME | ATTR_CTIME_SET)) {
1675                 /* For truncate and utimes sending attributes to OSTs, setting
1676                  * mtime/atime to the past will be performed under PW [0:EOF]
1677                  * extent lock (new_size:EOF for truncate).  It may seem
1678                  * excessive to send mtime/atime updates to OSTs when not
1679                  * setting times to past, but it is necessary due to possible
1680                  * time de-synchronization between MDT inode and OST objects */
1681                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1682         }
1683
1684         /* If the file was restored, it needs to set dirty flag.
1685          *
1686          * We've already sent MDS_DATA_MODIFIED flag in
1687          * ll_md_setattr() for truncate. However, the MDT refuses to
1688          * set the HS_DIRTY flag on released files, so we have to set
1689          * it again if the file has been restored. Please check how
1690          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1691          *
1692          * Please notice that if the file is not released, the previous
1693          * MDS_DATA_MODIFIED has taken effect and usually
1694          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1695          * This way we can save an RPC for common open + trunc
1696          * operation. */
1697         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1698                 struct hsm_state_set hss = {
1699                         .hss_valid = HSS_SETMASK,
1700                         .hss_setmask = HS_DIRTY,
1701                 };
1702                 int rc2;
1703
1704                 rc2 = ll_hsm_state_set(inode, &hss);
1705                 /* truncate and write can happen at the same time, so that
1706                  * the file can be set modified even though the file is not
1707                  * restored from released state, and ll_hsm_state_set() is
1708                  * not applicable for the file, and rc2 < 0 is normal in this
1709                  * case. */
1710                 if (rc2 < 0)
1711                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1712                                PFID(ll_inode2fid(inode)), rc2);
1713         }
1714
1715         EXIT;
1716 out:
1717         if (op_data != NULL)
1718                 ll_finish_md_op_data(op_data);
1719
1720         if (S_ISREG(inode->i_mode)) {
1721                 mutex_lock(&inode->i_mutex);
1722                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1723                         inode_dio_wait(inode);
1724         }
1725
1726         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1727                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1728
1729         return rc;
1730 }
1731
1732 int ll_setattr(struct dentry *de, struct iattr *attr)
1733 {
1734         int mode = de->d_inode->i_mode;
1735
1736         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1737                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1738                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1739
1740         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1741                                (ATTR_SIZE|ATTR_MODE)) &&
1742             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1743              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1744               !(attr->ia_mode & S_ISGID))))
1745                 attr->ia_valid |= ATTR_FORCE;
1746
1747         if ((attr->ia_valid & ATTR_MODE) &&
1748             (mode & S_ISUID) &&
1749             !(attr->ia_mode & S_ISUID) &&
1750             !(attr->ia_valid & ATTR_KILL_SUID))
1751                 attr->ia_valid |= ATTR_KILL_SUID;
1752
1753         if ((attr->ia_valid & ATTR_MODE) &&
1754             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1755             !(attr->ia_mode & S_ISGID) &&
1756             !(attr->ia_valid & ATTR_KILL_SGID))
1757                 attr->ia_valid |= ATTR_KILL_SGID;
1758
1759         return ll_setattr_raw(de, attr, false);
1760 }
1761
1762 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1763                        __u64 max_age, __u32 flags)
1764 {
1765         struct ll_sb_info *sbi = ll_s2sbi(sb);
1766         struct obd_statfs obd_osfs;
1767         int rc;
1768         ENTRY;
1769
1770         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1771         if (rc) {
1772                 CERROR("md_statfs fails: rc = %d\n", rc);
1773                 RETURN(rc);
1774         }
1775
1776         osfs->os_type = sb->s_magic;
1777
1778         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1779                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1780
1781         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1782                 flags |= OBD_STATFS_NODELAY;
1783
1784         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1785         if (rc) {
1786                 CERROR("obd_statfs fails: rc = %d\n", rc);
1787                 RETURN(rc);
1788         }
1789
1790         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1791                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1792                obd_osfs.os_files);
1793
1794         osfs->os_bsize = obd_osfs.os_bsize;
1795         osfs->os_blocks = obd_osfs.os_blocks;
1796         osfs->os_bfree = obd_osfs.os_bfree;
1797         osfs->os_bavail = obd_osfs.os_bavail;
1798
1799         /* If we don't have as many objects free on the OST as inodes
1800          * on the MDS, we reduce the total number of inodes to
1801          * compensate, so that the "inodes in use" number is correct.
1802          */
1803         if (obd_osfs.os_ffree < osfs->os_ffree) {
1804                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1805                         obd_osfs.os_ffree;
1806                 osfs->os_ffree = obd_osfs.os_ffree;
1807         }
1808
1809         RETURN(rc);
1810 }
1811 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1812 {
1813         struct super_block *sb = de->d_sb;
1814         struct obd_statfs osfs;
1815         __u64 fsid = huge_encode_dev(sb->s_dev);
1816         int rc;
1817
1818         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1819         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1820
1821         /* Some amount of caching on the client is allowed */
1822         rc = ll_statfs_internal(sb, &osfs,
1823                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1824                                 0);
1825         if (rc)
1826                 return rc;
1827
1828         statfs_unpack(sfs, &osfs);
1829
1830         /* We need to downshift for all 32-bit kernels, because we can't
1831          * tell if the kernel is being called via sys_statfs64() or not.
1832          * Stop before overflowing f_bsize - in which case it is better
1833          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1834         if (sizeof(long) < 8) {
1835                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1836                         sfs->f_bsize <<= 1;
1837
1838                         osfs.os_blocks >>= 1;
1839                         osfs.os_bfree >>= 1;
1840                         osfs.os_bavail >>= 1;
1841                 }
1842         }
1843
1844         sfs->f_blocks = osfs.os_blocks;
1845         sfs->f_bfree = osfs.os_bfree;
1846         sfs->f_bavail = osfs.os_bavail;
1847         sfs->f_fsid.val[0] = (__u32)fsid;
1848         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1849         return 0;
1850 }
1851
1852 void ll_inode_size_lock(struct inode *inode)
1853 {
1854         struct ll_inode_info *lli;
1855
1856         LASSERT(!S_ISDIR(inode->i_mode));
1857
1858         lli = ll_i2info(inode);
1859         mutex_lock(&lli->lli_size_mutex);
1860 }
1861
1862 void ll_inode_size_unlock(struct inode *inode)
1863 {
1864         struct ll_inode_info *lli;
1865
1866         lli = ll_i2info(inode);
1867         mutex_unlock(&lli->lli_size_mutex);
1868 }
1869
1870 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1871 {
1872         struct ll_inode_info *lli = ll_i2info(inode);
1873         struct mdt_body *body = md->body;
1874         struct ll_sb_info *sbi = ll_i2sbi(inode);
1875
1876         if (body->mbo_valid & OBD_MD_FLEASIZE)
1877                 cl_file_inode_init(inode, md);
1878
1879         if (S_ISDIR(inode->i_mode)) {
1880                 int     rc;
1881
1882                 rc = ll_update_lsm_md(inode, md);
1883                 if (rc != 0)
1884                         return rc;
1885         }
1886
1887         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1888                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1889                         ll_update_remote_perm(inode, md->remote_perm);
1890         }
1891 #ifdef CONFIG_FS_POSIX_ACL
1892         else if (body->mbo_valid & OBD_MD_FLACL) {
1893                 spin_lock(&lli->lli_lock);
1894                 if (lli->lli_posix_acl)
1895                         posix_acl_release(lli->lli_posix_acl);
1896                 lli->lli_posix_acl = md->posix_acl;
1897                 spin_unlock(&lli->lli_lock);
1898         }
1899 #endif
1900         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1901                                         sbi->ll_flags & LL_SBI_32BIT_API);
1902         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1903
1904         if (body->mbo_valid & OBD_MD_FLATIME) {
1905                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1906                         LTIME_S(inode->i_atime) = body->mbo_atime;
1907                 lli->lli_atime = body->mbo_atime;
1908         }
1909
1910         if (body->mbo_valid & OBD_MD_FLMTIME) {
1911                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1912                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1913                                "to "LPU64"\n", inode->i_ino,
1914                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1915                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1916                 }
1917                 lli->lli_mtime = body->mbo_mtime;
1918         }
1919
1920         if (body->mbo_valid & OBD_MD_FLCTIME) {
1921                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1922                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1923                 lli->lli_ctime = body->mbo_ctime;
1924         }
1925
1926         if (body->mbo_valid & OBD_MD_FLMODE)
1927                 inode->i_mode = (inode->i_mode & S_IFMT) |
1928                                 (body->mbo_mode & ~S_IFMT);
1929
1930         if (body->mbo_valid & OBD_MD_FLTYPE)
1931                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1932                                 (body->mbo_mode & S_IFMT);
1933
1934         LASSERT(inode->i_mode != 0);
1935         if (S_ISREG(inode->i_mode))
1936                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1937                                        LL_MAX_BLKSIZE_BITS);
1938         else
1939                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1940
1941         if (body->mbo_valid & OBD_MD_FLUID)
1942                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1943         if (body->mbo_valid & OBD_MD_FLGID)
1944                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1945         if (body->mbo_valid & OBD_MD_FLFLAGS)
1946                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1947         if (body->mbo_valid & OBD_MD_FLNLINK)
1948                 set_nlink(inode, body->mbo_nlink);
1949         if (body->mbo_valid & OBD_MD_FLRDEV)
1950                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1951
1952         if (body->mbo_valid & OBD_MD_FLID) {
1953                 /* FID shouldn't be changed! */
1954                 if (fid_is_sane(&lli->lli_fid)) {
1955                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1956                                  "Trying to change FID "DFID
1957                                  " to the "DFID", inode "DFID"(%p)\n",
1958                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1959                                  PFID(ll_inode2fid(inode)), inode);
1960                 } else {
1961                         lli->lli_fid = body->mbo_fid1;
1962                 }
1963         }
1964
1965         LASSERT(fid_seq(&lli->lli_fid) != 0);
1966
1967         if (body->mbo_valid & OBD_MD_FLSIZE) {
1968                 i_size_write(inode, body->mbo_size);
1969
1970                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1971                        PFID(ll_inode2fid(inode)),
1972                        (unsigned long long)body->mbo_size);
1973
1974                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1975                         inode->i_blocks = body->mbo_blocks;
1976         }
1977
1978         if (body->mbo_valid & OBD_MD_TSTATE) {
1979                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1980                  * clear it when done to ensure to start again
1981                  * glimpsing updated attrs
1982                  */
1983                 if (body->mbo_t_state & MS_RESTORE)
1984                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1985                 else
1986                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1987         }
1988
1989         return 0;
1990 }
1991
1992 int ll_read_inode2(struct inode *inode, void *opaque)
1993 {
1994         struct lustre_md *md = opaque;
1995         struct ll_inode_info *lli = ll_i2info(inode);
1996         int     rc;
1997         ENTRY;
1998
1999         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2000                PFID(&lli->lli_fid), inode);
2001
2002         /* Core attributes from the MDS first.  This is a new inode, and
2003          * the VFS doesn't zero times in the core inode so we have to do
2004          * it ourselves.  They will be overwritten by either MDS or OST
2005          * attributes - we just need to make sure they aren't newer. */
2006         LTIME_S(inode->i_mtime) = 0;
2007         LTIME_S(inode->i_atime) = 0;
2008         LTIME_S(inode->i_ctime) = 0;
2009         inode->i_rdev = 0;
2010         rc = ll_update_inode(inode, md);
2011         if (rc != 0)
2012                 RETURN(rc);
2013
2014         /* OIDEBUG(inode); */
2015
2016 #ifdef HAVE_BACKING_DEV_INFO
2017         /* initializing backing dev info. */
2018         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2019 #endif
2020         if (S_ISREG(inode->i_mode)) {
2021                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2022                 inode->i_op = &ll_file_inode_operations;
2023                 inode->i_fop = sbi->ll_fop;
2024                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2025                 EXIT;
2026         } else if (S_ISDIR(inode->i_mode)) {
2027                 inode->i_op = &ll_dir_inode_operations;
2028                 inode->i_fop = &ll_dir_operations;
2029                 EXIT;
2030         } else if (S_ISLNK(inode->i_mode)) {
2031                 inode->i_op = &ll_fast_symlink_inode_operations;
2032                 EXIT;
2033         } else {
2034                 inode->i_op = &ll_special_inode_operations;
2035
2036                 init_special_inode(inode, inode->i_mode,
2037                                    inode->i_rdev);
2038
2039                 EXIT;
2040         }
2041
2042         return 0;
2043 }
2044
2045 void ll_delete_inode(struct inode *inode)
2046 {
2047         struct ll_inode_info *lli = ll_i2info(inode);
2048         ENTRY;
2049
2050         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2051                 /* It is last chance to write out dirty pages,
2052                  * otherwise we may lose data while umount */
2053                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2054
2055         truncate_inode_pages_final(&inode->i_data);
2056
2057         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2058                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2059                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2060
2061 #ifdef HAVE_SBOPS_EVICT_INODE
2062         ll_clear_inode(inode);
2063 #endif
2064         clear_inode(inode);
2065
2066         EXIT;
2067 }
2068
2069 int ll_iocontrol(struct inode *inode, struct file *file,
2070                  unsigned int cmd, unsigned long arg)
2071 {
2072         struct ll_sb_info *sbi = ll_i2sbi(inode);
2073         struct ptlrpc_request *req = NULL;
2074         int rc, flags = 0;
2075         ENTRY;
2076
2077         switch(cmd) {
2078         case FSFILT_IOC_GETFLAGS: {
2079                 struct mdt_body *body;
2080                 struct md_op_data *op_data;
2081
2082                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2083                                              0, 0, LUSTRE_OPC_ANY,
2084                                              NULL);
2085                 if (IS_ERR(op_data))
2086                         RETURN(PTR_ERR(op_data));
2087
2088                 op_data->op_valid = OBD_MD_FLFLAGS;
2089                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2090                 ll_finish_md_op_data(op_data);
2091                 if (rc) {
2092                         CERROR("%s: failure inode "DFID": rc = %d\n",
2093                                sbi->ll_md_exp->exp_obd->obd_name,
2094                                PFID(ll_inode2fid(inode)), rc);
2095                         RETURN(-abs(rc));
2096                 }
2097
2098                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2099
2100                 flags = body->mbo_flags;
2101
2102                 ptlrpc_req_finished(req);
2103
2104                 RETURN(put_user(flags, (int __user *)arg));
2105         }
2106         case FSFILT_IOC_SETFLAGS: {
2107                 struct iattr *attr;
2108                 struct md_op_data *op_data;
2109                 struct cl_object *obj;
2110
2111                 if (get_user(flags, (int __user *)arg))
2112                         RETURN(-EFAULT);
2113
2114                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2115                                              LUSTRE_OPC_ANY, NULL);
2116                 if (IS_ERR(op_data))
2117                         RETURN(PTR_ERR(op_data));
2118
2119                 op_data->op_attr_flags = flags;
2120                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2121                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2122                 ll_finish_md_op_data(op_data);
2123                 ptlrpc_req_finished(req);
2124                 if (rc)
2125                         RETURN(rc);
2126
2127                 inode->i_flags = ll_ext_to_inode_flags(flags);
2128
2129                 obj = ll_i2info(inode)->lli_clob;
2130                 if (obj == NULL)
2131                         RETURN(0);
2132
2133                 OBD_ALLOC_PTR(attr);
2134                 if (attr == NULL)
2135                         RETURN(-ENOMEM);
2136
2137                 attr->ia_valid = ATTR_ATTR_FLAG;
2138                 rc = cl_setattr_ost(obj, attr, flags);
2139
2140                 OBD_FREE_PTR(attr);
2141                 RETURN(rc);
2142         }
2143         default:
2144                 RETURN(-ENOSYS);
2145         }
2146
2147         RETURN(0);
2148 }
2149
2150 int ll_flush_ctx(struct inode *inode)
2151 {
2152         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2153
2154         CDEBUG(D_SEC, "flush context for user %d\n",
2155                from_kuid(&init_user_ns, current_uid()));
2156
2157         obd_set_info_async(NULL, sbi->ll_md_exp,
2158                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2159                            0, NULL, NULL);
2160         obd_set_info_async(NULL, sbi->ll_dt_exp,
2161                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2162                            0, NULL, NULL);
2163         return 0;
2164 }
2165
2166 /* umount -f client means force down, don't save state */
2167 void ll_umount_begin(struct super_block *sb)
2168 {
2169         struct ll_sb_info *sbi = ll_s2sbi(sb);
2170         struct obd_device *obd;
2171         struct obd_ioctl_data *ioc_data;
2172         ENTRY;
2173
2174         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2175                sb->s_count, atomic_read(&sb->s_active));
2176
2177         obd = class_exp2obd(sbi->ll_md_exp);
2178         if (obd == NULL) {
2179                 CERROR("Invalid MDC connection handle "LPX64"\n",
2180                        sbi->ll_md_exp->exp_handle.h_cookie);
2181                 EXIT;
2182                 return;
2183         }
2184         obd->obd_force = 1;
2185
2186         obd = class_exp2obd(sbi->ll_dt_exp);
2187         if (obd == NULL) {
2188                 CERROR("Invalid LOV connection handle "LPX64"\n",
2189                        sbi->ll_dt_exp->exp_handle.h_cookie);
2190                 EXIT;
2191                 return;
2192         }
2193         obd->obd_force = 1;
2194
2195         OBD_ALLOC_PTR(ioc_data);
2196         if (ioc_data) {
2197                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2198                               sizeof *ioc_data, ioc_data, NULL);
2199
2200                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2201                               sizeof *ioc_data, ioc_data, NULL);
2202
2203                 OBD_FREE_PTR(ioc_data);
2204         }
2205
2206         /* Really, we'd like to wait until there are no requests outstanding,
2207          * and then continue.  For now, we just invalidate the requests,
2208          * schedule() and sleep one second if needed, and hope.
2209          */
2210         schedule();
2211         EXIT;
2212 }
2213
2214 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2215 {
2216         struct ll_sb_info *sbi = ll_s2sbi(sb);
2217         char *profilenm = get_profile_name(sb);
2218         int err;
2219         __u32 read_only;
2220
2221         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2222                 read_only = *flags & MS_RDONLY;
2223                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2224                                          sizeof(KEY_READ_ONLY),
2225                                          KEY_READ_ONLY, sizeof(read_only),
2226                                          &read_only, NULL);
2227                 if (err) {
2228                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2229                                       profilenm, read_only ?
2230                                       "read-only" : "read-write", err);
2231                         return err;
2232                 }
2233
2234                 if (read_only)
2235                         sb->s_flags |= MS_RDONLY;
2236                 else
2237                         sb->s_flags &= ~MS_RDONLY;
2238
2239                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2240                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2241                                       read_only ?  "read-only" : "read-write");
2242         }
2243         return 0;
2244 }
2245
2246 /**
2247  * Cleanup the open handle that is cached on MDT-side.
2248  *
2249  * For open case, the client side open handling thread may hit error
2250  * after the MDT grant the open. Under such case, the client should
2251  * send close RPC to the MDT as cleanup; otherwise, the open handle
2252  * on the MDT will be leaked there until the client umount or evicted.
2253  *
2254  * In further, if someone unlinked the file, because the open handle
2255  * holds the reference on such file/object, then it will block the
2256  * subsequent threads that want to locate such object via FID.
2257  *
2258  * \param[in] sb        super block for this file-system
2259  * \param[in] open_req  pointer to the original open request
2260  */
2261 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2262 {
2263         struct mdt_body                 *body;
2264         struct md_op_data               *op_data;
2265         struct ptlrpc_request           *close_req = NULL;
2266         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2267         ENTRY;
2268
2269         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2270         OBD_ALLOC_PTR(op_data);
2271         if (op_data == NULL) {
2272                 CWARN("%s: cannot allocate op_data to release open handle for "
2273                       DFID"\n",
2274                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2275
2276                 RETURN_EXIT;
2277         }
2278
2279         op_data->op_fid1 = body->mbo_fid1;
2280         op_data->op_handle = body->mbo_handle;
2281         op_data->op_mod_time = cfs_time_current_sec();
2282         md_close(exp, op_data, NULL, &close_req);
2283         ptlrpc_req_finished(close_req);
2284         ll_finish_md_op_data(op_data);
2285
2286         EXIT;
2287 }
2288
2289 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2290                   struct super_block *sb, struct lookup_intent *it)
2291 {
2292         struct ll_sb_info *sbi = NULL;
2293         struct lustre_md md = { NULL };
2294         int rc;
2295         ENTRY;
2296
2297         LASSERT(*inode || sb);
2298         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2299         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2300                               sbi->ll_md_exp, &md);
2301         if (rc != 0)
2302                 GOTO(cleanup, rc);
2303
2304         if (*inode) {
2305                 rc = ll_update_inode(*inode, &md);
2306                 if (rc != 0)
2307                         GOTO(out, rc);
2308         } else {
2309                 LASSERT(sb != NULL);
2310
2311                 /*
2312                  * At this point server returns to client's same fid as client
2313                  * generated for creating. So using ->fid1 is okay here.
2314                  */
2315                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2316
2317                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2318                                              sbi->ll_flags & LL_SBI_32BIT_API),
2319                                  &md);
2320                 if (IS_ERR(*inode)) {
2321 #ifdef CONFIG_FS_POSIX_ACL
2322                         if (md.posix_acl) {
2323                                 posix_acl_release(md.posix_acl);
2324                                 md.posix_acl = NULL;
2325                         }
2326 #endif
2327                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2328                         *inode = NULL;
2329                         CERROR("new_inode -fatal: rc %d\n", rc);
2330                         GOTO(out, rc);
2331                 }
2332         }
2333
2334         /* Handling piggyback layout lock.
2335          * Layout lock can be piggybacked by getattr and open request.
2336          * The lsm can be applied to inode only if it comes with a layout lock
2337          * otherwise correct layout may be overwritten, for example:
2338          * 1. proc1: mdt returns a lsm but not granting layout
2339          * 2. layout was changed by another client
2340          * 3. proc2: refresh layout and layout lock granted
2341          * 4. proc1: to apply a stale layout */
2342         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2343                 struct lustre_handle lockh;
2344                 struct ldlm_lock *lock;
2345
2346                 lockh.cookie = it->d.lustre.it_lock_handle;
2347                 lock = ldlm_handle2lock(&lockh);
2348                 LASSERT(lock != NULL);
2349                 if (ldlm_has_layout(lock)) {
2350                         struct cl_object_conf conf;
2351
2352                         memset(&conf, 0, sizeof(conf));
2353                         conf.coc_opc = OBJECT_CONF_SET;
2354                         conf.coc_inode = *inode;
2355                         conf.coc_lock = lock;
2356                         conf.u.coc_layout = md.layout;
2357                         (void)ll_layout_conf(*inode, &conf);
2358                 }
2359                 LDLM_LOCK_PUT(lock);
2360         }
2361
2362         GOTO(out, rc = 0);
2363
2364 out:
2365         md_free_lustre_md(sbi->ll_md_exp, &md);
2366
2367 cleanup:
2368         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2369                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2370
2371         return rc;
2372 }
2373
2374 int ll_obd_statfs(struct inode *inode, void __user *arg)
2375 {
2376         struct ll_sb_info *sbi = NULL;
2377         struct obd_export *exp;
2378         char *buf = NULL;
2379         struct obd_ioctl_data *data = NULL;
2380         __u32 type;
2381         int len = 0, rc;
2382
2383         if (!inode || !(sbi = ll_i2sbi(inode)))
2384                 GOTO(out_statfs, rc = -EINVAL);
2385
2386         rc = obd_ioctl_getdata(&buf, &len, arg);
2387         if (rc)
2388                 GOTO(out_statfs, rc);
2389
2390         data = (void*)buf;
2391         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2392             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2393                 GOTO(out_statfs, rc = -EINVAL);
2394
2395         if (data->ioc_inllen1 != sizeof(__u32) ||
2396             data->ioc_inllen2 != sizeof(__u32) ||
2397             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2398             data->ioc_plen2 != sizeof(struct obd_uuid))
2399                 GOTO(out_statfs, rc = -EINVAL);
2400
2401         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2402         if (type & LL_STATFS_LMV)
2403                 exp = sbi->ll_md_exp;
2404         else if (type & LL_STATFS_LOV)
2405                 exp = sbi->ll_dt_exp;
2406         else
2407                 GOTO(out_statfs, rc = -ENODEV);
2408
2409         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2410         if (rc)
2411                 GOTO(out_statfs, rc);
2412 out_statfs:
2413         if (buf)
2414                 obd_ioctl_freedata(buf, len);
2415         return rc;
2416 }
2417
2418 int ll_process_config(struct lustre_cfg *lcfg)
2419 {
2420         struct super_block *sb;
2421         unsigned long x;
2422         int rc = 0;
2423         char *ptr;
2424
2425         /* The instance name contains the sb: lustre-client-aacfe000 */
2426         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2427         if (!ptr || !*(++ptr))
2428                 return -EINVAL;
2429         if (sscanf(ptr, "%lx", &x) != 1)
2430                 return -EINVAL;
2431         sb = (struct super_block *)x;
2432         /* This better be a real Lustre superblock! */
2433         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2434
2435         /* Note we have not called client_common_fill_super yet, so
2436            proc fns must be able to handle that! */
2437         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2438                                       lcfg, sb);
2439         if (rc > 0)
2440                 rc = 0;
2441         return rc;
2442 }
2443
2444 /* this function prepares md_op_data hint for passing it down to MD stack. */
2445 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2446                                       struct inode *i1, struct inode *i2,
2447                                       const char *name, size_t namelen,
2448                                       __u32 mode, __u32 opc, void *data)
2449 {
2450         LASSERT(i1 != NULL);
2451
2452         if (name == NULL) {
2453                 /* Do not reuse namelen for something else. */
2454                 if (namelen != 0)
2455                         return ERR_PTR(-EINVAL);
2456         } else {
2457                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2458                         return ERR_PTR(-ENAMETOOLONG);
2459
2460                 if (!lu_name_is_valid_2(name, namelen))
2461                         return ERR_PTR(-EINVAL);
2462         }
2463
2464         if (op_data == NULL)
2465                 OBD_ALLOC_PTR(op_data);
2466
2467         if (op_data == NULL)
2468                 return ERR_PTR(-ENOMEM);
2469
2470         ll_i2gids(op_data->op_suppgids, i1, i2);
2471         op_data->op_fid1 = *ll_inode2fid(i1);
2472         op_data->op_default_stripe_offset = -1;
2473         if (S_ISDIR(i1->i_mode)) {
2474                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2475                 if (opc == LUSTRE_OPC_MKDIR)
2476                         op_data->op_default_stripe_offset =
2477                                    ll_i2info(i1)->lli_def_stripe_offset;
2478         }
2479
2480         if (i2) {
2481                 op_data->op_fid2 = *ll_inode2fid(i2);
2482                 if (S_ISDIR(i2->i_mode))
2483                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2484         } else {
2485                 fid_zero(&op_data->op_fid2);
2486         }
2487
2488         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2489                 op_data->op_cli_flags |= CLI_HASH64;
2490
2491         if (ll_need_32bit_api(ll_i2sbi(i1)))
2492                 op_data->op_cli_flags |= CLI_API32;
2493
2494         op_data->op_name = name;
2495         op_data->op_namelen = namelen;
2496         op_data->op_mode = mode;
2497         op_data->op_mod_time = cfs_time_current_sec();
2498         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2499         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2500         op_data->op_cap = cfs_curproc_cap_pack();
2501         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2502              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2503                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2504         } else {
2505                 op_data->op_mds = 0;
2506         }
2507         op_data->op_data = data;
2508
2509         return op_data;
2510 }
2511
2512 void ll_finish_md_op_data(struct md_op_data *op_data)
2513 {
2514         OBD_FREE_PTR(op_data);
2515 }
2516
2517 #ifdef HAVE_SUPEROPS_USE_DENTRY
2518 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2519 #else
2520 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2521 #endif
2522 {
2523         struct ll_sb_info *sbi;
2524
2525 #ifdef HAVE_SUPEROPS_USE_DENTRY
2526         LASSERT((seq != NULL) && (dentry != NULL));
2527         sbi = ll_s2sbi(dentry->d_sb);
2528 #else
2529         LASSERT((seq != NULL) && (vfs != NULL));
2530         sbi = ll_s2sbi(vfs->mnt_sb);
2531 #endif
2532
2533         if (sbi->ll_flags & LL_SBI_NOLCK)
2534                 seq_puts(seq, ",nolock");
2535
2536         if (sbi->ll_flags & LL_SBI_FLOCK)
2537                 seq_puts(seq, ",flock");
2538
2539         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2540                 seq_puts(seq, ",localflock");
2541
2542         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2543                 seq_puts(seq, ",user_xattr");
2544
2545         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2546                 seq_puts(seq, ",lazystatfs");
2547
2548         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2549                 seq_puts(seq, ",user_fid2path");
2550
2551         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2552                 seq_puts(seq, ",always_ping");
2553
2554         RETURN(0);
2555 }
2556
2557 /**
2558  * Get obd name by cmd, and copy out to user space
2559  */
2560 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2561 {
2562         struct ll_sb_info *sbi = ll_i2sbi(inode);
2563         struct obd_device *obd;
2564         ENTRY;
2565
2566         if (cmd == OBD_IOC_GETDTNAME)
2567                 obd = class_exp2obd(sbi->ll_dt_exp);
2568         else if (cmd == OBD_IOC_GETMDNAME)
2569                 obd = class_exp2obd(sbi->ll_md_exp);
2570         else
2571                 RETURN(-EINVAL);
2572
2573         if (!obd)
2574                 RETURN(-ENOENT);
2575
2576         if (copy_to_user((void __user *)arg, obd->obd_name,
2577                          strlen(obd->obd_name) + 1))
2578                 RETURN(-EFAULT);
2579
2580         RETURN(0);
2581 }
2582
2583 /**
2584  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2585  * fsname will be returned in this buffer; otherwise, a static buffer will be
2586  * used to store the fsname and returned to caller.
2587  */
2588 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2589 {
2590         static char fsname_static[MTI_NAME_MAXLEN];
2591         struct lustre_sb_info *lsi = s2lsi(sb);
2592         char *ptr;
2593         int len;
2594
2595         if (buf == NULL) {
2596                 /* this means the caller wants to use static buffer
2597                  * and it doesn't care about race. Usually this is
2598                  * in error reporting path */
2599                 buf = fsname_static;
2600                 buflen = sizeof(fsname_static);
2601         }
2602
2603         len = strlen(lsi->lsi_lmd->lmd_profile);
2604         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2605         if (ptr && (strcmp(ptr, "-client") == 0))
2606                 len -= 7;
2607
2608         if (unlikely(len >= buflen))
2609                 len = buflen - 1;
2610         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2611         buf[len] = '\0';
2612
2613         return buf;
2614 }
2615
2616 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2617 {
2618         char *path = NULL;
2619
2620         struct path p;
2621
2622         p.dentry = dentry;
2623         p.mnt = current->fs->root.mnt;
2624         path_get(&p);
2625         path = d_path(&p, buf, bufsize);
2626         path_put(&p);
2627         return path;
2628 }
2629
2630 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2631 {
2632         char *buf, *path = NULL;
2633         struct dentry *dentry = NULL;
2634         struct inode *inode = page->mapping->host;
2635
2636         /* this can be called inside spin lock so use GFP_ATOMIC. */
2637         buf = (char *)__get_free_page(GFP_ATOMIC);
2638         if (buf != NULL) {
2639                 dentry = d_find_alias(page->mapping->host);
2640                 if (dentry != NULL)
2641                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2642         }
2643
2644         CDEBUG(D_WARNING,
2645                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2646                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2647                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2648                PFID(ll_inode2fid(inode)),
2649                (path && !IS_ERR(path)) ? path : "", ioret);
2650
2651         if (dentry != NULL)
2652                 dput(dentry);
2653
2654         if (buf != NULL)
2655                 free_page((unsigned long)buf);
2656 }
2657
2658 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2659                         struct lov_user_md **kbuf)
2660 {
2661         struct lov_user_md      lum;
2662         ssize_t                 lum_size;
2663         ENTRY;
2664
2665         if (copy_from_user(&lum, md, sizeof(lum)))
2666                 RETURN(-EFAULT);
2667
2668         lum_size = ll_lov_user_md_size(&lum);
2669         if (lum_size < 0)
2670                 RETURN(lum_size);
2671
2672         OBD_ALLOC(*kbuf, lum_size);
2673         if (*kbuf == NULL)
2674                 RETURN(-ENOMEM);
2675
2676         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2677                 OBD_FREE(*kbuf, lum_size);
2678                 RETURN(-EFAULT);
2679         }
2680
2681         RETURN(lum_size);
2682 }
2683
2684 /*
2685  * Compute llite root squash state after a change of root squash
2686  * configuration setting or add/remove of a lnet nid
2687  */
2688 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2689 {
2690         struct root_squash_info *squash = &sbi->ll_squash;
2691         int i;
2692         bool matched;
2693         lnet_process_id_t id;
2694
2695         /* Update norootsquash flag */
2696         down_write(&squash->rsi_sem);
2697         if (list_empty(&squash->rsi_nosquash_nids))
2698                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2699         else {
2700                 /* Do not apply root squash as soon as one of our NIDs is
2701                  * in the nosquash_nids list */
2702                 matched = false;
2703                 i = 0;
2704                 while (LNetGetId(i++, &id) != -ENOENT) {
2705                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2706                                 continue;
2707                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2708                                 matched = true;
2709                                 break;
2710                         }
2711                 }
2712                 if (matched)
2713                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2714                 else
2715                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2716         }
2717         up_write(&squash->rsi_sem);
2718 }
2719
2720 /**
2721  * Parse linkea content to extract information about a given hardlink
2722  *
2723  * \param[in]   ldata      - Initialized linkea data
2724  * \param[in]   linkno     - Link identifier
2725  * \param[out]  parent_fid - The entry's parent FID
2726  * \param[out]  ln         - Entry name destination buffer
2727  *
2728  * \retval 0 on success
2729  * \retval Appropriate negative error code on failure
2730  */
2731 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2732                             struct lu_fid *parent_fid, struct lu_name *ln)
2733 {
2734         unsigned int    idx;
2735         int             rc;
2736         ENTRY;
2737
2738         rc = linkea_init(ldata);
2739         if (rc < 0)
2740                 RETURN(rc);
2741
2742         if (linkno >= ldata->ld_leh->leh_reccount)
2743                 /* beyond last link */
2744                 RETURN(-ENODATA);
2745
2746         linkea_first_entry(ldata);
2747         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2748                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2749                                     parent_fid);
2750                 if (idx == linkno)
2751                         break;
2752
2753                 linkea_next_entry(ldata);
2754         }
2755
2756         if (idx < linkno)
2757                 RETURN(-ENODATA);
2758
2759         RETURN(0);
2760 }
2761
2762 /**
2763  * Get parent FID and name of an identified link. Operation is performed for
2764  * a given link number, letting the caller iterate over linkno to list one or
2765  * all links of an entry.
2766  *
2767  * \param[in]     file - File descriptor against which to perform the operation
2768  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2769  *                       on and the available size. It is eventually filled with
2770  *                       the requested information or left untouched on error
2771  *
2772  * \retval - 0 on success
2773  * \retval - Appropriate negative error code on failure
2774  */
2775 int ll_getparent(struct file *file, struct getparent __user *arg)
2776 {
2777         struct dentry           *dentry = file->f_path.dentry;
2778         struct inode            *inode = dentry->d_inode;
2779         struct linkea_data      *ldata;
2780         struct lu_buf            buf = LU_BUF_NULL;
2781         struct lu_name           ln;
2782         struct lu_fid            parent_fid;
2783         __u32                    linkno;
2784         __u32                    name_size;
2785         int                      rc;
2786
2787         ENTRY;
2788
2789         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2790             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2791                 RETURN(-EPERM);
2792
2793         if (get_user(name_size, &arg->gp_name_size))
2794                 RETURN(-EFAULT);
2795
2796         if (get_user(linkno, &arg->gp_linkno))
2797                 RETURN(-EFAULT);
2798
2799         if (name_size > PATH_MAX)
2800                 RETURN(-EINVAL);
2801
2802         OBD_ALLOC(ldata, sizeof(*ldata));
2803         if (ldata == NULL)
2804                 RETURN(-ENOMEM);
2805
2806         rc = linkea_data_new(ldata, &buf);
2807         if (rc < 0)
2808                 GOTO(ldata_free, rc);
2809
2810         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2811         if (rc < 0)
2812                 GOTO(lb_free, rc);
2813
2814         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2815         if (rc < 0)
2816                 GOTO(lb_free, rc);
2817
2818         if (ln.ln_namelen >= name_size)
2819                 GOTO(lb_free, rc = -EOVERFLOW);
2820
2821         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2822                 GOTO(lb_free, rc = -EFAULT);
2823
2824         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2825                 GOTO(lb_free, rc = -EFAULT);
2826
2827         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2828                 GOTO(lb_free, rc = -EFAULT);
2829
2830 lb_free:
2831         lu_buf_free(&buf);
2832 ldata_free:
2833         OBD_FREE(ldata, sizeof(*ldata));
2834
2835         RETURN(rc);
2836 }